1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References
  8. Supporting Information

Viral hepatitis is the leading cause of liver disease worldwide and can be caused by several agents, including hepatitis A (HAV), B (HBV), and C (HCV) virus. We employed multiplexed protein immune assays to identify biomarker signatures of viral hepatitis in order to define unique and common responses for three different acute viral infections of the liver. We performed multianalyte profiling, measuring the concentrations of 182 serum proteins obtained from acute HAV- (18), HBV- (18), and HCV-infected (28) individuals, recruited as part of a hospital-based surveillance program in Cairo, Egypt. Virus-specific biomarker signatures were identified and validation was performed using a unique patient population. A core signature of 46 plasma proteins was commonly modulated in all three infections, as compared to healthy controls. Principle component analysis (PCA) revealed a host response based upon 34 proteins, which could distinguish HCV patients from HAV- and HBV-infected individuals or healthy controls. When HAV and HBV groups were compared directly, 34 differentially expressed serum proteins allowed the separation of these two patient groups. A validation study was performed on an additional 111 patients, confirming the relevance of our initial findings, and defining the 17 analytes that reproducibly segregated the patient populations. Conclusions: This combined discovery and biomarker validation approach revealed a previously unrecognized virus-specific induction of host proteins. The identification of hepatitis virus specific signatures provides a foundation for functional studies and the identification of potential correlates of viral clearance. (Hepatology 2014;59:1273-1282)


acute hepatitis C


alpha 1 antitrypsin


alanine aminotransferase


analysis of variance




brain-derived neurotrophic factor


chemiluminescent immunoassay




Epstein-Barr virus


false discovery rate


hepatitis A virus


hepatitis B virus


hepatitis C virus


intercellular adhesion molecule 1




immunoglobulin M




IFN-γ-inducible protein 10


lowest detectable dose


multianalyte profiling


monocyte chemotactic protein 4


macrophage colony-stimulating factor


macrophage-inflammatory protein


open reading frame


pulmonary and activation-regulated chemokine


principle component analysis


reverse-transcriptase polymerase chain reaction




transforming growth factor alpha


tumor necrosis factor




upper limit of normal

The liver functions primarily as a metabolic organ and is in constant exposure to dietary and microbial antigens.[1] This requires a state of relative immune tolerance, which evolutionarily has been exploited by pathogens such as hepatitis viruses, bacteria, and parasites.[1] Three hepatotropic viruses, hepatitis A, B, and C (HAV, HBV, and HCV, respectively) all directly infect hepatocytes, triggering similar acute disease in symptomatic patients despite displaying differing modalities of infection and levels of pathogenicity.

HAV is a positive-strand RNA virus and a member of the Picornaviridae family, and infection results in a self-limiting acute disease with a considerable percentage of adult patients developing clinical symptoms of fever or jaundice.[2, 3] Partly as a result of a successful vaccine, relatively little is known about immune clearance of HAV, although recent studies in chimpanzees suggested a role for CD4+ T cells.[4] Earlier studies demonstrated that viral loads peak 2 weeks post-transmission, with a half-life (t1/2) of ∼9 days for viral populations in the liver, and are undetectable by weeks 6-8. Notably, clinical symptoms develop after week 4 and are coincident with the emergence of a humoral response to viral proteins.[3, 5]

HBV is an enveloped circular DNA virus belonging to the Hepadnaviridae family. It directly enters the nucleus of infected hepatocytes to initiate its replication cycle,[6] peak viral load occurs at 7-8 weeks post-transmission, and the reported t1/2 is 2-3 days.[6, 7] Similar to HAV, HBV elicits a limited type I interferon (IFN) response in the liver,[8, 9] and clearance is believed to be mediated by CD4+ and CD8+ T cells.[10] Approximately 5%-10% of acutely infected adults will progress to chronic HBV infection. Although a prophylactic vaccine exists, HBV remains a leading worldwide public health problem, with ∼350 million chronically infected persons and >4 million acute clinical cases per year.

HCV is an enveloped positive-strand RNA virus and is classified among the Flaviviridae family. In contrast to HAV and HBV, it results in chronic infection in >60% of cases, and no vaccine currently exists. HCV also has an early peak in viral load, however it is unique among the three viruses in its rapid turnover and induction of a strong type I IFN response in the liver.[11, 12] Studies in humans and chimpanzees have highlighted the importance of the adaptive immune response for achieving spontaneous clearance of the virus,[13] although neither cellular nor humoral immunity is believed to be fully protective against secondary infection.[14, 15]

A direct comparison of acute viral hepatitis infections in humans has not been previously performed from the perspective of the host response. This is, in part, a result of the limited incident of disease for HAV as well as the low percentage of symptomatic acute HCV patients in the United States and Europe. Beginning in 2002, we established a hospital-based surveillance program to monitor acute symptomatic viral hepatitis in Cairo, Egypt.[16] This program was established to characterize the current localized HCV epidemic, which resulted from iatrogenic transmissions during the mass treatment parenteral antischistosomiasis (PAT) campaigns in the 1970s.[17] In the context of this program, we identified patients monoinfected with either HAV, HBV, or HCV. To identify the inflammatory signatures that define the host response to these different infectious agents, we performed multianalyte profiling (MAP) using a panel of 182 chemiluminescent immunoassay (CLIA)-validated immunoassays. We report on a core signature profile that was common for acute infection of the liver by three different hepatotropic viruses, as well as serum protein biomarker signatures for each virus and host interaction. To identify patterns of virus-specific protein responses, we utilized principle component analysis (PCA), in combination with a novel statistical tool (the Projection Score),[18] that aimed to determine the optimal subset of features for distinguishing patient groups. A collection of feature subsets were obtained by ranking all analytes according to their adjusted P values from an analysis of variance (ANOVA), contrasting hepatitis subtypes and successively eliminating the analytes that showed the highest adjusted P values. The projection score for such a subset is a measure of the information content captured by PCA, and the optimal subset of features is defined as the one that provides the most informative PCA representation of the samples in the data set. Thus, by comparing the information content of variable subsets resulting from different cut-off levels for the adjusted P value, the projection score provides an objective, statistically well-founded way to establish an optimal cutoff. Notably, projection score analysis may represent a general mechanism for enabling statistical analysis of biomarker information to optimize the variable subsets in the context of discovery-based research. Subsequent to the identification of host response signatures, our core set of biomarkers was validated using an independent patient population. These results will provide a solid foundation for mechanistic studies and may provide insights into different hepatotropic viral disease pathogenesis.

Patients and Methods

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References
  8. Supporting Information
Patient, Cohort, and Study Design

Study 1 included 18 patients with HAV, 18 with HBV, 28 with HCV, and 15 healthy controls (Supporting Table 1), and the validation population, study 2, included 20 patients with HAV, 20 with HBV, and 71 with HCV (Supporting Table 2). All patients were recruited from two “fever hospitals” specialized in infectious diseases in Cairo, Egypt.[16] Inclusion criteria were characteristic clinical findings including fever or jaundice as well as an elevated alanine aminotransferase (ALT) ≥3 times the upper limit of normal (ULN; Supporting Tables 1 and 2). Acute HCV was identified by positive HCV-RNA (reverse-transcriptase polymerase chain reaction; RT-PCR) with either negative or positive anti-HCV antibody (Ab; Innotest HCV Ab IV; Innogenetics N.V., Ghent, Belgium). Patients with negative anti-HCV Ab and positive HCV-RNA were considered as acute hepatitis C (AHC) cases. Patients with positive anti-HCV Ab and positive HCV-RNA were considered AHC cases only if ALT values were >10 times ULN and there was a recent history of possible high-risk exposure to HCV (e.g., surgical procedure). Acute HBV was defined by a positive immunoglobulin M (IgM) anti-hepatitis B virus core (Corzyme M; ribosomal DNA; Abbott Laboratories, Abbott Park, IL) and circulating levels of hepatitis B surface antigen (Auszyme Monoclonal; third-generation enzyme immunoassay; Abbott Laboratories). Acute hepatitis A infection was defined as positive anti-HAV IgM Ab. Using RT-PCR for HEV-RNA (in-house assay using open reading frame ORF1 and ORF2 primers) and serological testing, including anti–Epstein-Barr virus (EBV) IgM (ETI-EBV-M reverse P001605; Dia Sorin, Vercelle, Italy), anti-cytomegalovirus (CMV) IgM (AXSYM system-CMV-IgM; Abbott Laboratories, Wiesbaden, Delknheim, Germany), and anti-Toxoplasma IgM (AXSYM system-Toxo-IgM; Abbott Laboratories, Germany), EBV, CMV, toxoplasmosis, and hepatitis E infection were all used as exclusion criteria. Serum samples analyzed in this study were collected with a median of 7-10 days postsymptoms (Supporting Table 1 and 2). All patients were treatment naïve because this was their first recorded incidence of hepatitis.

Table 1. Protein Analytes That Distinguish HAV, HBV, and HCV Patients From Healthy Controls
AnalyteP ValueQ ValueGroup
  1. The 24 most differential analytes are shown, indicating the P and Q values (ANOVA). Analytes were quantified using MAP and are listed in the order of statistical significance. Data shown are from the patients recruited as part of study 1. To identify the patient group(s) associated with the protein analyte, Dunn's post-test was performed.

  2. Abbreviations: HB-EGF, heparin-binding EGF-like growth factor; EGF, epidermal growth factor; TECK, thymus-expressed chemokine; VCAM-1, vascular cell adhesion molecule 1; NrCAM, neuronal cell adhesion molecule; AXL, AXL receptor tyrosine kinase; MIG, monokine-induced by gamma interferon.

ALT4.56 × 10−256.65 × 10−23ABC
PARC3.83 × 10−212.8 × 10−19C
HB-EGF9.8 × 10−184.77 × 10−16AB
Thrombomodulin4.51 × 10−171.64 × 10−15AB
Complement 31.07 × 10−153.12 × 10−14C
α1-AT1.38 × 10−153.35 × 10−14C
EGF2.1 × 10−154.38 × 10−14ABC
M-CSF1.03 × 10−131.89 × 10−12ABC
IP-102.32 × 10−133.77 × 10−12ABC
TECK4.97 × 10−137.26 × 10−12C
MCP-47.56 × 10−131 × 10−11AB
ICAM-12.21 × 10−122.7 × 10−11C
VCAM-13.49 × 10−123.92 × 10−11ABC
NrCAM4.08 × 10−114.25 × 10−10ABC
Apo H4.82 × 10−114.69 × 10−10C
IgM9.81 × 10−118.42 × 10−10ABC
AXL7.28 × 10−105.9 × 10−9AB
IgA1.96 × 10−91.51 × 10−8C
MIG2.49 × 10−91.82 × 10−8ABC
E-selectin1.8 × 10−81.24 × 10−7ABC
TTR1.8 × 10−81.24 × 10−7ABC
TNF RII3.18 × 10−82.09 × 10−7ABC
MIP-1α4.54 × 10−82.84 × 10−8AB
Apo D9.71 × 10−85.8 × 10−8C
Table 2. Protein Analytes That Distinguish HAV, HBV, and HCV Patients
Study 1Study 2
AnalyteP ValueQ ValueGroupAnalyteP ValueQ ValueGroup
  1. The 34 most differential analytes are shown, indicating the P and Q values (ANOVA). Analytes were quantified using MAP and are listed in the order of statistical significance, as determined from the analysis of data from the patients recruited as part of study 1. These same analytes were then examined in an independent cohort (study 2). To identify the patient group(s) associated with the protein analyte, Dunn's post-test was performed.

  2. Abbreviations: HB-EGF, heparin-binding epidermal growth factor (EGF)-like growth factor; vWF, von Willebrand factor; AXL, AXL receptor tyrosine kinase; VEGF, vascular endothelial growth factor; THP, Tamm-Horsfall protein; HGF, hepatocyte growth factor; PPP, protein serine/threonine phosphatases; TIMP-1, tissue inhibitor metallopeptidase inhibitor 1.

PARC7.55 × 10-241.05 × 10−21CIgM1.28 × 10−131.45 × 10−12C
HB-EGF4.03 × 10−202.8 × 10−18CApo B7.57 × 10−86.44 × 10−7B
Thrombomodulin1.15 × 10−134.88 × 10−12CM-CSF1.83 × 10−71.24 × 10−6C
α1-AT1.74 × 10−134.88 × 10−12CCD5L1.19 × 10−56.75 × 10−5B
Complement 31.75 × 10−134.88 × 10−12CComplement 33.2 × 10−41.5 × 10−3C
IgA5.14 × 10−111.19 × 10−9Cα1-AT4.6 × 10−31.9 × 10−2C
IgM8.17 × 10−111.62 × 10−9CApo D5.05 × 10−31.9 × 10−2C
MCP-41.28 × 10−102.23 × 10−9CPPP8.93 × 10−32.9 × 10−2A
BDNF2.45 × 10−93.92 × 10−8CApo H9.6 × 10−32.9 × 10−2C
Apo H1.71 × 10−72.23 × 10−7CAXL1.1 × 10−23.3 × 10−2B
vWF1.76 × 10−72.23 × 10−6CBDNF1.46 × 10−23.8 × 10−2C
Prolactin5.8 × 10−76.72 × 10−6COsteopontin1.66 × 10−24.03 × 10−2C
C-reactive protein2.31 × 10−82.34 × 10−7CTHP2.57 × 10−25.8 × 10−2C
AXL2.35 × 10−62.34 × 10−5BvWF3.03 × 10−26.45 × 10−2B
VEGF2.69 × 10−62.36 × 10−5CVEGF4.2 × 10−28.45 × 10−2
Apo B2.72 × 10−62.36 × 10−5CHGF4.5 × 10−28.5 × 10−2A
TGF-α6.8 × 10−65.56 × 10−5CTIMP-15.0 × 10−29.2 × 10−2C
Osteopontin1.06 × 10−58.18 × 10−5CApo AI5.4 × 10−29.2 × 10−2
THP1.54 × 10−51.12 × 10−4CLipoprotein (a)7.7 × 10−21.2 × 10−1
Clusterin1.61 × 10−51.12 × 10−4CClusterin7.9 × 10−21.2 × 10−1
MIP-1α1.86 × 10−51.22 × 10−4CMCP-49.1 × 10−21.3 × 10−1
CD5L2.46 × 10−51.49 × 10−4BTNF RII1.1 × 10−11.6 × 10−1
M-CSF2.46 × 10−51.49 × 10−4CThrombomodulin1.6 × 10−12.1 × 10−1
HGF3.9 × 10−52.2 × 10−4CProlactin1.6 × 10−12.1 × 10−1
TNF RII4.5 × 10−52.5 × 10−4CPARC1.9 × 10−12.4 × 10−1
Apo D5.7 × 10−52.7 × 10−4CMIP-1α2.8 × 10−13.4 × 10−1
Apo A-I7.17 × 10−53.6 × 10−4CICAM-13.8 × 10−14.4 × 10−1
ICAM-17.8 × 10−53.8 × 10−4CCD40 ligand5.4 × 10−16.1 × 10−1
CD40 ligand7.99 × 10−53.8 × 10−4CIgA5.6 × 10−16.2 × 10−1
IL-189.13 × 10−54.2 × 10−4CIL-187.0 × 10−17.4 × 10−1
PPP1.2 × 10−45.6 × 10−4BC-reactive protein7.5 × 10−17.7 × 10−1
Lipoprotein (a)4.5 × 10−41.9 × 10−3CIL-69.8 × 10−19.8 × 10−1
TIMP-17.1 × 10−43.0 × 10−3CHB-EGF
IL-61.3 × 10−35.4 × 10−3CTGF-α

Healthy control serum from 15 anonymous donors was obtained from the local Cairo blood bank and confirmed to be negative for HAV, HBV, HCV, and human immunodeficiency virus. Protocols were reviewed and approved by local ethical committees and patients provided informed consent. The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki.


Sera were clarified by a high-speed centrifugation and analyzed using Luminex xMAP technology. Samples were measured by a diagnostic lab (Myriad Rules Based Medicine) for 182 molecules with all assays being CLIA certified (validated using guidelines set forth by the U.S. Clinical and Laboratory Standards Institute, Wayne, PA). Additional details on determining assay sensitivity are included in the Supporting Materials (Supporting Table 3).

Table 3. Protein Analytes That Distinguish HAV and HBV Patients
Study 1Study 2
AnalyteP ValueQ Value<AnalyteP valueQ value<
  1. The 34 most differential analytes are shown, indicating the P and Q values (t test). Analytes were quantified using MAP and are listed in the order of statistical significance, as determined from the analysis of data from the patients recruited as part of study 1. These same analytes were then examined in an independent cohort (study 2).

  2. Abbreviations: EGFR, epidermal growth factor receptor; AXL, AXL receptor tyrosine kinase; PPP, protein serine/threonine phosphatases; SAP, shingolipid activator protein; SOD1, superoxide dismutase 1; ANG, angiopoietin; TBG, thyroxine-binding globulin; TTR, thyroxine-binding prealbumin; ACE, angiotensin-converting enzyme; PLGF, placental growth factor; VCAM-1, vascular cell adhesion molecule 1; MIG, monokine-induced by gamma interferon; FRTN, ferritin.

IP-104.3 × 10−44.74 × 10−2A<BIgM1.5 × 10−52.5 × 10−4B<A
EGFR6.5 × 10−44.74 × 10−2B<ATBG7.32 × 10−86.22 × 10−7A<B
Betacellulin5.13 × 10−32.46 × 10−1B<AACE1.95 × 10−61.33 × 10−5A<B
AXL7.5 × 10−32.87 × 10−1A<BCD5L4.97 × 10−62.82 × 10−5A<B
PPP1.14 × 10−22.87 × 10−1A<BANG-21.01 × 10−44.4 × 10−4A<B
CD5L1.19 × 10−22.87 × 10−1A<BSAP1.04 × 10−44.4 × 10−4B<A
MIP-1β1.51 × 10−23.1 × 10−1A<BTenascin C1.44 × 10−35.2 × 10−3A<B
SAP2.47 × 10−24.4 × 10−1B<AMIP-3α1.54 × 10−35.26 × 10−3A<B
IgM3.26 × 10−24.4 × 10−1B<AIP-102.0 × 10−36.2 × 10−3A<B
Tenascin C3.45 × 10−24.4 × 10−1A<BAXL5.33 × 10−31.42 × 10−2A<B
MIG3.71 × 10−24.4 × 10−1A<BPPP5.43 × 10−31.4 × 10−2A<B
SOD13.72 × 10−24.4 × 10−1B<AMIP-1β9.64 × 10−32.34 × 10−2A<B
TNF RII4.57 × 10−24.8 × 10−1A<BVCAM-11.67 × 10−23.8 × 10−2A<B
MIF4.68 × 10−24.8 × 10−1B<ATTR4.44 × 10−28.79 × 10−2B<A
ANG-25.5 × 10−24.8 × 10−1EGF4.49 × 10−28.79 × 10−2A<B
MIP-3α5.5 × 10−24.9 × 10−1Haptoglobin4.65 × 10−28.79 × 10−2B<A
Haptoglobin6.2 × 10−25.3 × 10−1Cystatin C5.2 × 10−29.3 × 10−2
IL-1β7.2 × 10−25.7 × 10−1MIG5.7 × 10−29.8 × 10−2
TBG7.8 × 10−25.7 × 10−1E-selectin8.8 × 10−21.42 × 10−2
TTR8.6 × 10−25.7 × 10−1MIG1.07 × 10−11.58 × 10−1
Ferritin8.7 × 10−25.7 × 10−1Apo A-II1.09 × 10−11.58 × 10−1
ACE9.6 × 10−25.7 × 10−1PLGF1.1 × 10−12.1 × 10−1
PLGF1.0 × 10−15.7 × 10−1SOD11.2 × 10−11.7 × 10−1
M-CSF1.0 × 10−15.7 × 10−1Cancer antigen 19-91.3 × 10−11.7 × 10−1
Apo A-II1.1 × 10−15.7 × 10−1Eotaxin-11.45 × 10−11.7 × 10−1
β2 microglobulin1.1 × 10−15.7 × 10−1Apo E1.5 × 10−11.7 × 10−1
Cancer antigen 19-91.1 × 10−15.7 × 10−1FRTN1.5 × 10−11.7 × 10−1
Eotaxin-11.2 × 10−15.7 × 10−1IL-182.1 × 10−12.3 × 10−1
E-selectin1.2 × 10−15.7 × 10−1M-CSF2.8 × 10−13.0 × 10−1
Apo E1.2 × 10−15.7 × 10−1MIF3.2 × 10−13.4 × 10−1
VCAM-11.2 × 10−15.7 × 10−1TNF RII4.1 × 10−14.3 × 10−1
EGF1.3 × 10−15.8 × 10−1β2 microglobin7.2 × 10−17.2 × 10−1
Cystatin C1.3 × 10−15.8 × 10−1EGFR
IL-181.4 × 10−16.2 × 10−1Betacellulin
Statistical Analyses

PCA, agglomerative hierarchical clustering (based on mean linkage), and ANOVA testing were performed with Qlucore Omics Explorer (Qlucore, Lund, Sweden). The projection score was computed and used to quantify the information content for each of a collection of variable subsets obtained by ordering the analytes with respect to their false discovery rate (FDR)-adjusted ANOVA P values and successively excluding the analytes with the highest adjusted P values. By selecting the variable subset with the highest information content (i.e., with the highest projection score), we objectively established an optimally adjusted P value threshold to use as inclusion criterion for defined protein signatures.[18] Kruskal-Wallis' and Mann-Whitney's unpaired tests and Spearman's correlation calculations were performed using OmniViz (BioWisdom Ltd., Cambridge, UK) or Prism (GraphPad Software Inc., San Diego, CA). Data are reported as standard P values and FDR-adjusted P values (called Q values) to control for multiple testing.[19]


  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References
  8. Supporting Information
A Common Hepatic Viral Signature

To examine the signature of the host response to HAV, HBV, and HCV infection, we performed Luminex xMAP on sera of patients, isolated at the time of acute, symptomatic infection. To clean the data set, we removed 43 analytes that had a median value across all samples (study 1) that was below the lowest detectable dose (LDD). For the remaining 139 analytes (Supporting Table 3), individual patients that had a value below the LDD were assigned a value half that of the lowest value detected in the data set. To identify a common hepatic viral signature, we compared HAV-, HBV-, and HCV-infected patients with healthy controls using two complementary strategies. First, we performed one-way ANOVA, contrasting healthy donors and the combined set of infected patients. This approach revealed 46 analytes that significantly distinguished patients from noninfected controls (P < 0.05). We also performed direct two-way comparisons between each infected group and healthy controls using OmniViz software and then examined the lists for common differentially expressed analytes. The two different approaches revealed similar results, verifying the different analytic approaches and establishing a common hepatic signature of viral infection (Fig. 1). Not surprisingly, the induced proteins included molecules that are known to be released secondary to liver injury, as well as elevated levels of inducible cytokines and chemokines. Interestingly, we also observed a profound decrease in levels of apolipoproteins (Apos) and other serum proteins known to be synthesized by hepatocytes. These data are consistent with immune activation and viral-induced subversion of normal hepatocyte function.


Figure 1. Common hepatitis viral signature. The analytes commonly up- (red box) or down-regulated (green box) in HAV, HBV, and HCV patients, as compared to healthy controls.

Download figure to PowerPoint

Hierarchical Clustering and PCA Stratify Patients With Acute Hepatitis Infection

To identify potential analytes that stratify patient groups, we applied an ANOVA to each individual variable, contrasting all four groups of subjects (HAV, HBV, HCV, and healthy controls). Using an unbiased analytic approach, the least discriminatory variables were sequentially removed and hierarchical clustering was performed (Fig. 2A). For each variable subset obtained, we calculated the projection score (based on the first two principal components)[18] and retained the subset of variable analytes that gave the maximal projection score (Fig. 2B). This resulted in the identification of 25 analytes with Q values less than 2.09 × 10−7 (Fig. 2A; Table 1). Then, we projected the 25 analytes using PCA and visualized the groups using the first three principal components. Before applying PCA, each of the 25 analytes were centered to zero mean and scaled to unit variance. The PCA showed a clear separation of HCV patients from HAV and HBV patients and healthy controls (Fig. 2C). Interestingly, HAV and HBV patients could not be separated using this analytical strategy. The first principal component (vector 1) captured 37% of the variance and was driven by the differences between healthy and infected patients (Fig. 2D). In other words, the discriminatory proteins were identified for showing a distribution of A/B/C ≠ healthy (Fig. 2D). The second principal component (representing 25% of the variance) was driven by differences between HCV patients and the other groups (Fig. 2E), mostly showing a distribution of A/B ≠ healthy ≠ C. The third principal component (5% of the variance) was driven by proteins that showed a distribution of A/B ≠ C/healthy (Fig. 2F). Together, the first three principal components accounted for 67% of the variance present in the 25 analytes, implying that the three-dimensional representation of the patients shown in Fig. 2C accurately represents the analyte variation across the patient groups. To illustrate further how the PCA is calculated, the relative values for three analytes (alpha 1 antitrypsin [α1-AT], ALT, monocyte chemotactic protein 4 [MCP-4]) across the three dimensions are shown (Fig. 2C).


Figure 2. Comparison of HAV, HBV, HCV, and healthy serum samples by MAP. (A) Heat map showing hierarchical clustering of HAV-, HBV-, and HCV-infected patients and healthy controls based on 24 analytes measured in serum by MAP. (B) Projection score as determined by the number of included variables in a comparison of HAV-, HBV-, and HCV-infected patients and healthy controls. (C) PCA showing separation between HAV/HBV- and HCV-infected individuals and healthy controls based on 24 analytes measured in serum by MAP and determined by the projection score. (Each dot represents a single patient designated by the color code.) Relative values of α1-AT, ALT, and MCP-4 for each patient across the PCA space. (D-F) The relative contribution of each analyte to PCA vector 1 (D), vector 2 (E) and vector 3 (F) are shown. The direction of the bars is arbitrary and is not considered negative or positive, except in relation to the other analytes.

Download figure to PowerPoint

Direct Comparison of Hepatitis Patients

Although HCV patients were separable from the other study groups, it was surprising to observe such similar signatures in HAV and HBV patients. Because the inclusion of healthy donors introduces a bias in the analysis, we focused on differences that may exist between HAV, HBV, and HCV patients in the absence of a reference population. Again, we applied an ANOVA, contrasting the three patient subgroups, and successively excluded the least significant analytes. The projection score (based on two principal components, because the centroids of the three patient subgroups generically span a two-dimensional space) was used to determine a suitable inclusion threshold resulting in the most informative variable subset. In the absence of the healthy control group, ANOVA analysis revealed 34 analytes that differentiated HAV/HBV- from HCV-infected patients (Q < 5.4 × 10−3; Table 2). Sixteen of these analytes were previously identified in Fig. 2; however, removal of healthy controls resulted in the identification of an additional 18 analytes that were differentially expressed between HCV and HAV/HBV. PCA revealed a clear separation of these groups based on the 34 included analytes (Fig. 3A). Strikingly, the most common pattern observed was a higher serum concentration of analytes in HAV and HBV, as compared to HCV, patients. The exceptions were transforming growth factor alpha (TGF-α), thrombomodulin, brain-derived neurotrophic factor (BDNF), and intercellular adhesion molecule 1 (ICAM-1), which were higher in the HCV patient group. The contribution of each analyte to the three PCA vectors is shown in Supporting Fig. 1.


Figure 3. Comparison of HAV-, HBV-, and HCV-infected serum samples by MAP. (A) PCA showing full separation between HAV/HBV- and HCV-infected individuals in study 1 based on 34 analytes measured in serum by MAP and selected based on the projection score. (Each dot represents a single patient designated by the color code.) (B) Projection score as determined by the number of included variables in a comparison of HAV-, HBV-, and HCV-infected patients. (C) PCA showing full separation between HAV/HBV- and HCV-infected individuals in study 2 based on the 34 analytes that were identified in study 1. (Each dot represents a single patient designated by the color code.)

Download figure to PowerPoint

To validate the proteins identified as differentially expressed between hepatitis patient groups, we sampled an additional group of HAV, HBV, and HCV patients that were identified and recruited 1-5 years after the initial study (Supporting Table 2). These samples were processed in the same way as the initial study, and MAP Luminex assays were performed on plasma for the selected analytes. Using the 34 proteins identified from study 1 as the input variables (Fig. 2), we performed PCA on samples from study 2. Based on data from the first three principle components, we were able to demonstrate that HCV patients were clustered as well as separable from HAV and HBV patients (Fig. 3C). Whereas the PCA distance between the groups of patients in study 2 was less significant, we could identify a subset of 17 variables that remained differentially expressed (P < 0.05), verifying both the approach used and highlighting the importance of validation studies to confirm discovery-based biomarker results (Table 2). The individual values for the 17 protein biomarkers identified in both studies 1 and 2 are plotted in Supporting Fig. 3. Despite being significantly different across two independent studies, certain proteins (Complement 3; BDNF) displayed an inconsistent pattern between the two studies (Supporting Fig. 3). The identified biomarkers showed no correlations with viral load.

Two-Way Comparison of HAV Versus HBV-Infected Patients

Despite their close association in the multigroup analysis, we hypothesized that HAV- and HBV-infected patients had differentially expressed proteins that may be revealed by excluding HCV-infected patients from the analysis. To evaluate this possibility, we performed a t test, directly comparing HAV and HBV patients. PCA was applied, removing the least significant analytes, and projecting data from the 34 remaining variables onto the first three component axes (Fig. 4A). We again utilized the projection score (based on the first principal component, because the discrimination of the two groups contrasted with the t test is expected to occur along a single dimension) to determine the optimal cutoff for the number of analytes used for discriminating the groups. The P and Q values for each of the 34 individual analytes are reported in Table 3.


Figure 4. Direct comparison of HAV- and HBV-infected serum samples. (A) PCA showing separation between HAV- and HBV-infected individuals in study 1 based on 34 analytes measured in serum by MAP and selected based on the projection score. (Each dot represents a single patient designated by the color code.) (B) Projection score as determined by the number of included variables in a comparison of HAV- and HBV-infected patients. (C) PCA showing separation between HAV- and HBV-infected individuals in study 2 based on the 34 analytes that were identified in study 1. (Each dot represents a single patient designated by the color code.)

Download figure to PowerPoint

To assess the robustness of the discriminatory analytes, we applied a similar approach as described above to the assessment of data from study 2. Utilizing the 34 identified analytes as input values for the PCA, we were able to separate HAV from HBV patients (Fig. 4C). Notably, the two patient groups were better stratified than in the initial study, and 16 of the analytes were significantly different using univariate tests (P < 0.05).


  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References
  8. Supporting Information

In this study, we have taken advantage of a unique Egyptian viral hepatitis surveillance program[16] to characterize and compare the host immune response to acute HAV, HBV, and HCV infection. To our knowledge, this is the first direct comparison of the human inflammatory response to acute infection by HAV, HBV, and HCV. An important parameter to consider when directly comparing acute viral hepatic infections is the kinetic differences in the onset of symptoms and disease pathogenesis. For HAV, the average incubation period is 30 days, after which most people show nonspecific constitutional symptoms, followed by gastrointestinal symptoms and development of jaundice.[3] Subsequent to HBV infection, the incubation period is reported to be between 2 and 3 months, although virus can be detected in the circulation within 1 month.[6] Clinical symptoms, such as fatigue, nausea, and jaundice, are reported to occur in 1 in 3 of HBV-infected individuals. By contrast, acute HCV infection is mostly asymptomatic, with virus detectable in the blood after 7-21 days.[20] The clinical syndrome, when present, is believed to occur between 2 and 12 weeks after infection.[20] Despite these kinetic differences, we were able to identify a common hepatic viral signature (Fig. 1). The majority of molecules that showed reduced levels in all three infections were liver-associated proteins, with seven of those (α1-AT, Apo A1, Apo B, Apo D, Apo H, Apo J, and lipoprotein a) having even lower levels in HCV patients. Depressed hepatocyte function may be a reflection of host pathways being modulated by the virus or, alternatively, a consequence of liver inflammation. Consistent with the latter hypothesis, proteins commonly up-regulated in all three infections included several cytokines, chemokines, and growth factors. Antiviral host effectors, IFN-γ, tumor necrosis factor (TNF)-α, and interleukin (IL)-18, were up-regulated in all three infections, suggesting that the different clinical outcomes are a reflection of the individual viruses' relationship with the host.

To explore the unique pathogen-induced host response, we focused on the variable protein signatures within the data set. Interestingly, many of the analytes that were differentially expressed in HAV and HBV, compared to HCV, across both study groups are liver-associated proteins, growth factors, or humoral response proteins. Ab responses, which are considered important for halting the spread of viral infection, are induced earlier in HAV and HBV infections than in HCV.[4, 6] IgM was found to be consistently higher in HAV than HBV patients, who had higher levels than HCV. Other immune modulators that showed consistently lower levels in HCV patients, compared to HAV and HBV, are macrophage colony-stimulating factor (M-CSF) and osteopontin. A set of liver-associated proteins also demonstrated differential expression patterns between infection types: Apo B, Apo D, Apo H, and α1-AT were all significantly lower in HCV infections, as compared to HAV- and HBV-infected patients. The association between HCV and host Apos has been previously well recognized.[26] These results suggest that HCV interferes to a much greater extent with normal liver function, as compared to HAV or HBV.

Upon initial assessment, analyte expression in HAV and HBV patients appeared similar when examined in the same analysis as HCV patients. However, when HAV and HBV patients were compared directly, we could detect consistent differences. The combination of PCA and the projection score allowed us to identify a group of analytes that conventional statistical approaches would have overlooked, which were then validated in study 2. Some of the most differential proteins between these two groups, consistently at higher levels in HBV infection, are immune proteins, such as IFN-γ-inducible protein 10 (IP-10), CD5L, macrophage-inflammatory protein (MIP)−1β, and MIP-3α. Others that showed higher levels in HAV patients are proteins produced by the liver, such as transthyretin (TTR) and haptoglobin. Additional studies will be required to assess whether these differences are directly caused by the virus or the resultant inflammatory response.

Although the strengths of our study lie in the use of a clinically validated proteomic tool and access to unique patient populations, there are limitations to discovery-based research. To address and overcome these potential caveats, we employed novel statistical tools, such as the projection score, which, utilized in combination with PCA, dramatically improved the identification of protein biomarker combinations. To test our approach and validate the initially identified discriminatory proteins, we undertook a full validation study on distinct patient groups. Despite a large time gap between the two studies, 50% of our identified proteins remained statistically significant and differential between patient groups. However, highlighting the danger of solitary cohorts for biomarker discovery was the observation that the most differentiating analyte in study 1, pulmonary and activation-regulated chemokine (PARC), showed no differences between patient groups in study 2.

In summary, we have demonstrated the ability to use a discovery-based approach and clustering algorithms for the identification and validation of discriminatory biomarkers. Additionally, we highlight the definition of protein signatures that are commonly induced by viral infection of the liver, as well as those analytes that are differentially expressed during the unique viral/host detente. These results provide a solid foundation for a larger effort to understand the pathogenesis of hepatic viral infection.


  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References
  8. Supporting Information

The authors thank Stéphanie Thomas and Nehal Al-Saqqa for their project management support and the patient volunteers that participated in our study. The authors also thank the Center for Human Immunology, Institut Pasteur, for their support of the experimental work.


  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References
  8. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Patients and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References
  8. Supporting Information

Additional Supporting Information may be found in the online version of this article.

hep26901-sup-0001-suppfigs.pdf1142KSupplementary Information Figure S1-S6.
hep26901-sup-0002-supptables.doc339KSupplementary Information Table S1-S3.
hep26901-sup-0003-suppinfo.doc27KSupplementary Information

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.